An Efficient Method for Partitioning Initial Chromosomes in the Island Model Genetic Algorithm

Abstract

In this thesis I propose a proper method of partitioning initial population in Island Model GAs. Island Models are a popular and efficient way to implement a genetic algorithm on a parallel machine. In an Island Model each machine maintains its own subpopulation using a genetic algorithm for search. In the basic algorithm of parallel GAs, all populations are shuffled randomly across the different "islands". That is, the initial population is partitioned in random, although we can control this unknown factor on purpose. I designed two methods of partitioning initial population which considers the distances among population rather than random partitioning. Experimental results show that one of the methods brings a considerable improvement in finding a global optima.